Memory devices (e.g., standalone dynamic random access memory (DRAM), embedded DRAM) that store data in a capacitor in a memory cell need to be timely refreshed to restore a charge stored in the capacitor that leaks away in a data retention stage of the memory cell. Due to a non-uniform leakage distribution of memory cells in a memory device, some cells, called “weak” cells, have a higher leakage current than an average leakage current of other cells, called “normal” cells, and fail to retain data at the same refresh rate as that of other cells. In order to eliminate read errors, the weak cells found during testing of the memory device are repaired. Moreover, throughout the life time of the memory device, some cells that perform normally under the refresh rate set initially during the testing of the memory device may eventually degrade and become weak cells that need to be repaired.
One approach to repair the weak cells is to select a refresh rate sufficiently high to compensate for both the average leakage current in the average cells and the higher leakage current in the weak cells. Unfortunately, this approach causes increase of overall refresh current of the memory device and significant increase of power consumption, and reduces a bandwidth for accessing the memory device.
Another approach to repair the weak cells involves identifying weak cells and repairing them by assigning spare rows or columns to replace the rows or columns containing the weak cells. For this approach, the refresh rate can be set in accordance with the average leakage current in the average cells. However, it is too costly to repair, for example, one weak cell in a memory block with a whole spare row or column, and, if a weak cell again appears in the spare row or column, another spare row or column needs to be allocated for repairing the weak cell in the spare row or column, and therefore, double resources are used to fix one weak cell in the memory block.
For weak cells that appear throughout the life time of the memory device, the above-mentioned approaches are even more uneconomical because increasing the overall refresh rate causes a significant percentage of increase in power consumption in view of the percentage of power consumption of the weak cells, and setting aside double spare rows or columns for each possible occurrence of weak cell results in a significant percentage increase in area in view of the percentage of area of the weak cells.
Like reference symbols in the various drawings indicate like elements.